[GERONIMODEVTOOLS-608] Publish with GEP takes minutes, while deploy takes seconds

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.2.0
• Fix Version/s: 2.2.1, 2.1.7, 3.0
• Component/s: eclipse-plugin
• Environment: GEP 2.2 installed in Eclipse (galileo-SR1).
• Assignee: Xuan Dai
• Reporter: Boes
• Created: 01/Apr/10 07:41
• Updated: 17/Jan/11 09:11
• Resolved: 14/Oct/10 07:22

Description

Publishing an Enterprise Application (EAR) with GEP to Geronimo takes much more time then deploying the same EAR to Geronimo using the console. See http://n3.nabble.com/Publish-with-GEP-takes-minutes-while-deploy-takes-10-seconds-td684484.html

After doing some research I found that EAR's that have dependencies in geronimo-application.xml take a lot of time to publish. This is caused by the very inefficient implementation of the reorderModules method in the org.apache.geronimo.st.core.internal.DependencyHelper class.

I installed a GEP development environment and put on tracing. It tested an EAR with 3 WAR's in it. In the EAR's geronimo-application.xml I added dependency tags for 4 libraries. Tracing shows that this results in parsing the geronimo-application.xml 1092 times. In code this means that a call to DependencyHelper.getEnvironment is made 1092 times.

Another inefficient part in the reorderModules method is that the call to DeploymentUtils.isInstalledModule is repeatedly made for the same module. In the test I found it was called at least 3 times for each dependent library. These calls result in a request to Geronimo and take almost a second each.

The best way to solve this bug is to redesign the reordering process and make it work in a way that both DependencyHelper.getEnvironment and DeploymentUtils.isInstalledModule are not called more often then needed. I wonder why DeploymentUtils.isInstalledModule is called at all. Dependencies are defined in the xml. I can't imagine why it would matter if a module is installed in Geronimo or not.

What I did to fix this problem and make GEP workable again for me is:

• made sure the DeploymentUtils.isInstalledModule is never called twice for the same module. Once the DeploymentUtils.isInstalledModule is called for a certain module, I add this module to a list. Before a next call to DeploymentUtils.isInstalledModule I verify if the call has been made before. If so, it doesn't need to be called again.
• reduced the number of times the xml is parsed. I put the results of the DependencyHelper.getEnvironment for a certain module in a hashmap. Next time the DependencyHelper.getEnvironment is called for the same module, I returns the result stored in the hashmap.

After these two modifications GEP is up to speed again.

Removal of the call to DeploymentUtils.isInstalledModule makes GEP even faster. Just as fast as deployment using the Geronimo console. For me this works, but I don't know what the impact is for other users. In the attached code I left it the way it was.